In [1]:
import cv2
from cv2 import VideoCapture
from sklearn.svm import SVC
from os import path
import os
import numpy as np
from matplotlib import pyplot as plt
%matplotlib inline
In [2]:
n_pos = len(os.listdir('target/'))
n_neg = len(os.listdir('error/'))
dataset = [('target/' + f, 1) for f in os.listdir('target/')] + [('error/' + f, 0) for f in os.listdir('error/')]
for i in np.random.choice(os.listdir('target/'), size=(n_neg - n_pos), replace=True):
    dataset.append(('target/' + i, 1))
dataset = np.array(dataset)
N = len(dataset)
val_id = np.random.choice(range(N), N // 10, replace=False)
val_set = dataset[val_id]
train_set = dataset[~val_id]
In [3]:
img = cv2.imread('target/00020240.jpg')
plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
Out[3]:
<matplotlib.image.AxesImage at 0x7fef6323e310>
In [4]:
def feature(image_name):
    feat = []
    if isinstance(image_name, basestring):
        img = cv2.imread(image_name)
    else:
        img = image_name
    feat.append(img.mean(axis=(0, 1)))
    for x in np.hsplit(img, 3):    # x is (1080, 640, 3)
        for y in np.vsplit(x, 3):  # y is (360, 640, 3)
            feat.append(y.mean(axis=(0, 1)))
    
    return np.array(feat).flatten()
In [5]:
def featurize(dataset):
    X = []
    y = []
    for img, tag in dataset:
        try:
            X.append(feature(img))
            y.append(int(tag))
        except:
            print img, 'is not an image'
            pass
    X = np.array(X)
    y = np.array(y)
    return X, y
In [6]:
X, y = featurize(train_set)
VX, Vy = featurize(val_set)
np.mean(y), np.mean(Vy)

target/test.py is not an image
target/.DS_Store is not an image
target/test.py is not an image
target/.DS_Store is not an image
target/.DS_Store is not an image
target/.DS_Store is not an image
target/test.py is not an image
Out[6]:
(0.5174825174825175, 0.47120418848167539)
In [19]:
classifier = SVC(probability=True)
classifier.fit(X, y)
Out[19]:
SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0,
  decision_function_shape=None, degree=3, gamma='auto', kernel='rbf',
  max_iter=-1, probability=True, random_state=None, shrinking=True,
  tol=0.001, verbose=False)
In [20]:
predy = classifier.predict(VX)
print 'total accuracy', np.mean(predy == Vy)
print 'false positive', np.mean((Vy == 0) & (predy == 1))
print 'false negative', np.mean((Vy == 1) & (predy == 0))

total accuracy 0.970331588133
false positive 0.0
false negative 0.0296684118674
In [44]:
def video_feature(name):
    vid = VideoCapture(name)
    frames = []
    feat = []
    i = 0
    while True:
        i += 1
        if i < 10:
            continue
        i = 0
        ret, img = vid.read()
        if not ret:
            break
        feat.append(feature(img))
        frames.append(img)
    return np.array(feat), np.array(frames)
In [45]:
TX, frame = video_feature('001e4073d0aaef120fb35e9175370297.mp4')
In [46]:
TX.shape
Out[46]:
(3532, 30)
In [47]:
Ty = classifier.predict(TX)
In [48]:
Ty.sum()
Out[48]:
0
In [59]:
z = []
for i in range(len(TX) - 1):
    z.append(np.linalg.norm(TX[i] - TX[i + 1]))
    if z[-1] > 50:
        print i, z[-1]
        plt.imshow(cv2.cvtColor(frame[i], cv2.COLOR_BGR2RGB))
        plt.show()
plt.plot(z)

119 320.848323015

209 52.1775430171

292 332.885010695

418 339.173193511

708 247.398446461

760 181.723112969

1188 215.566330869

1298 201.615005327

1383 60.1065765339

1401 55.4218259593

1402 84.8028753433

1403 88.5961462539

1404 127.501885673

1405 197.042132545

1406 186.103144418

1573 60.0217518863

1574 68.8841408726

1575 81.4956192019

1576 136.938631539

1577 180.134794892

1578 165.889817259

1941 193.000600806

2044 197.828110632

2257 255.759956635

2375 321.87023667

2517 84.5496178128

2518 100.82588597

2519 105.839270191

2520 171.071592335

2521 173.229773911

3005 97.4353212374

3006 108.032735363

3007 101.811156201

3008 175.919389062

3009 187.175204294

3217 157.06331986

3359 177.137442249

3518 279.13576883
Out[59]:
[<matplotlib.lines.Line2D at 0x7fef49cbdc90>]
In [82]:
def find_continue_group(ary):
    group_start = -1
    for a, b in zip(ary[:-1], ary[1:]):
        if b - a > 1:
            if group_start > 0 and group_len > 1:
                print group_start, group_len
            group_start = b
            group_len = 0
        elif b - a == 1 and group_start > 0:
            group_len += 1
In [79]:
np.where(np.array(z) > 50)[0]
Out[79]:
array([ 119,  209,  292,  418,  708,  760, 1188, 1298, 1383, 1401, 1402,
       1403, 1404, 1405, 1406, 1573, 1574, 1575, 1576, 1577, 1578, 1941,
       2044, 2257, 2375, 2517, 2518, 2519, 2520, 2521, 3005, 3006, 3007,
       3008, 3009, 3217, 3359, 3518])
In [83]:
find_continue_group(np.where(np.array(z) > 50)[0])

1401 5
1573 5
2517 4
3005 4
In [88]:
for i in range(2517 + 20, 3005 - 20, 100):
    plt.imshow(cv2.cvtColor(frame[i], cv2.COLOR_BGR2RGB))
    plt.show()
In [86]:
for i in range(1401 + 20, 1573 - 20, 10):
    plt.imshow(cv2.cvtColor(frame[i], cv2.COLOR_BGR2RGB))
    plt.show()
In [ ]: